Glacial inceptions: Past and future

Transcription

1 Atmosphere-Ocean ISSN: (Print) (Online) Journal homepage: Glacial inceptions: Past and future Lawrence A. Mysak To cite this article: Lawrence A. Mysak (2008) Glacial inceptions: Past and future, Atmosphere- Ocean, 46:3, , DOI: /ao To link to this article: Published online: 21 Nov Submit your article to this journal Article views: 388 View related articles Citing articles: 9 View citing articles Full Terms & Conditions of access and use can be found at

2 Glacial Inceptions: Past and Future Lawrence A. Mysak* Department of Atmospheric and Oceanic Sciences McGill University, 805 Sherbrooke Street W. Montreal QC H3A 2K6 [Original manuscript received 28 August 2007; accepted 4 February 2008] ABSTRACT The realistic simulation of northern hemisphere glacial inceptions, which occurred during the Quaternary period, has challenged scores of climate theoreticians and modellers. After reviewing the Milankovitch theory of glaciation, a number of earlier modelling studies of the last glacial inception (LGI) which have employed either high-resolution General Circulation Models (GCMs) or Earth system Models of Intermediate Complexity (EMICs) are described. The latter class of models has been developed over the past two decades in order to investigate the many interactions and feedbacks among the geophysical and biospheric components of the Earth system that take place over long time scales. Following a description of the McGill Paleoclimate Model (MPM) and other EMICs, some recent McGill simulations of the LGI in response to orbital (Milankovitch) and radiative (atmospheric CO 2 ) forcings are presented. Special attention is given to determining the relative roles of the ocean s thermohaline circulation, freshwater fluxes into the ocean, orography, cryospheric processes and vegetation dynamics during the inception phase. In particular, it is shown that with the vegetation-albedo feedback included in the model, the buildup of ice sheets over North America is larger than over Eurasia, in agreement with observations. This paper concludes with a discussion on the (possible) occurrence of the next glacial period. To address this issue, which has been inspired by recent publications of Berger and Loutre, MPM simulations of the climate for the next 100 kyr, forced by various prescribed atmospheric CO 2 levels, as well as the future insolation changes as calculated by the Berger algorithm, are presented. The influence of a near-term global warming scenario on glacial inception is also examined. If it is assumed that after such a warming scenario the concentration of CO 2 in the atmosphere returns to pre-industrial levels (in the range of ppm), then the MPM predicts that the next glacial would start at around 50 kyr after present, which is consistent with the results of Berger and Loutre. Finally, recent simulations of future glacial inceptions using the Potsdam EMIC which includes an atmosphereocean carbon cycle component are described. From one of these simulations in which 5000 GtC are released into the atmosphere due to human activities, it is concluded that the current interglacial will last for at least another half-million years because of the limited ability of the oceans to absorb such a large carbon release to the atmosphere. RÉSUMÉ [Traduit par la rédaction] La simulation réaliste des débuts des périodes glaciaires qui se sont produites durant le Quaternaire dans l hémisphère Nord a constitué un défi pour nombre de théoriciens et de modélisateurs du climat. Après un examen de la théorie de Milankovitch sur les glaciations, nous décrivons un certain nombre d études de modélisation du début de la dernière période glaciaire (DDPG) faites précédemment et qui ont utilisé des modèles de circulation générale (MCG) à haute résolution ou des modèles de système terrestre de complexité intermédiaire (EMIC). Cette dernière classe de modèles a été mise au point au cours des deux dernières décennies dans le but d étudier les nombreuses interactions et rétroactions se produisant entre les éléments géophysiques et biosphériques du système terrestre sur de grandes échelles de temps. Après une description du modèle paléoclimatique de McGill (MPM) et d autres EMIC, certaines simulations récentes de McGill du DDPG en réponse aux forçages orbital (Milankovitch) et radiatif (CO 2 atmosphérique) sont présentées. Nous nous efforçons en particulier de déterminer les rôles relatifs de la circulation thermohaline océanique, des flux d eau douce vers l océan, de l orographie, des processus cryosphériques et de la dynamique de la végétation durant la phase initiale. Nous montrons notamment qu avec la rétroaction végétation-albédo incluse dans le modèle, l accumulation de couches glaciaires sur l Amérique du Nord est plus grande que sur l Eurasie, conformément aux observations. Cet article se termine par une discussion sur la (possible) prochaine période glaciaire. Pour étudier cette question, qui s inscrit dans la foulée de publications récentes de Berger et Loutre, nous présentons des simulations du MPM du climat des prochaines années, qui sont basées sur un forçage par différents niveaux de CO 2 atmosphérique spécifiés de même que sur les variations futures de l insolation telles que This paper is dedicated to the memory of Stephen Mysak, *Author s lawrence.mysak@mcgill.ca Canadian Meteorological and Oceanographic Society

3 318 / Lawrence A. Mysak calculées au moyen de l algorithme de Berger. Nous examinons aussi l influence d un scénario de réchauffement de la planète dans un avenir rapproché sur l amorce d une période glaciaire. Si l on suppose qu après un tel scénario de réchauffement, la concentration en CO 2 atmosphérique revient aux niveaux préindustriels (dans intervalle ppm), alors le MPM prévoit que la prochaine glaciation commencerait dans environ ans, ce qui correspond aux résultats de Berger et Loutre. Finalement, nous décrivons des simulations récentes de futurs débuts de période glaciaire réalisées avec l EMIC de Potsdam, qui comporte une composante de cycle de carbone atmosphère-océan. L une de ces simulations, dans laquelle GtC sont libérées dans l atmosphère par les activités humaines, mène à la conclusion que l époque interglaciale actuelle durera au moins un autre demi-million d années, à cause de la capacité limitée des océans d absorber une aussi importante libération de carbone dans l atmosphère. 1 Introduction Over the past few million years, the evolution of the climate on orbital (Milankovitch) time scales has exhibited various quasi-periodic fluctuations, the most prominent of which are the relatively recent 100-kyr ice age cycles which can be seen in ice-sheet and deep-ocean sediment core records (Fig. 1). The alternation between relatively short, warm interglacials and long, cold glacials during the past half-million years is attributed to a complex set of processes that involve orbital forcing and internal interactions and feedbacks in the climate system (Ruddiman, 2001). The main goal of this paper is to present a review of various attempts to simulate the transition from an interglacial to a glacial period, which is commonly referred to as a glacial inception. Both past and future glacial inceptions will be considered. However, since the literature on glacial inceptions is rather vast, the focus will be on (1) the last glacial inception (LGI) at around 116 kyr BP (before present; here present is defined as 1950), which occurred during marine isotope stage (MIS) 5d, and (2) the occurrence of the next possible glacial inception. Further, as this paper represents a personal perspective on the topic, the simulations presented here are mainly from the Earth system modelling group at McGill University. Many of the earlier modelling studies on glacial inception have involved the use of atmospheric General Circulation Models (GCMs) run with a fixed seasonal cycle, a low-level summer insolation (i.e., incoming solar radiation at the top of the atmosphere) in northern boreal latitudes and a fixed radiative (atmospheric CO 2 ) forcing extending over a few decades or centuries (e.g., see Oglesby (1990), de Noblet et al. (1996), and the references therein). These simulations are called timeslice runs. However, since the main driver of the ice age cycles and a glacial inception in particular is the slow change in insolation that takes place over tens of thousands of years due to variations in the Earth s orbital parameters, recent modelling studies of these climate changes have employed Earth system Models of Intermediate Complexity (EMICs; Claussen et al., 2002), which are forced by temporally and spatially varying fields that are prescribed for these long time periods. Such EMIC simulations are called transient runs, wherein the different components of the climate system evolve and interact on a variety of time scales in response to the slowly changing forcing fields. Among the pioneers who developed EMICs to simulate glacial-interglacial cycles are the many investigators of the Louvain-la-Neuve group in Belgium (e.g., see Gallée et al., 1992; Loutre and Berger, 2000). However, it is interesting to note that the first use of the phrase intermediate complexity in describing a coupled global climate model seems to have occurred in Stocker et al. (1992). The EMIC simulations discussed in this review could be considered an extension of earlier transient runs of coupled atmosphere-ice sheet models, in which the atmosphere is represented by an energy balance model (e.g., see Tarasov and Peltier (1999) and the references therein). Since these twocomponent models do not include an interactive ocean, land surface, vegetation, or atmospheric moisture components (in contrast to most EMICs), they will not be discussed further in this review. With EMICs, it is possible to address many questions that cannot be answered using GCMs. For example, how and when did the last glacial start, over what period of time and where did the ice sheets grow substantially in the northern hemisphere? What are the roles of the different climate components in a glacial inception? Given that the last few interglacials (namely, those occurring during MIS 5e, 7 and 9) lasted only on the order of ten thousand years (Fig. 1), it is natural to ask whether the present interglacial, the 10-kyr Holocene (MIS 1), will end in the not-too-distant future. Over the past few decades, it has been forecasted (Kukla et al., 1972; Broecker, 1998) that the Holocene would indeed terminate soon. More recently, it has been proposed that had it not been for the slow anthropogenic emission of greenhouse gases thousands of years ago (due to early forest clearance for farming in Eurasia and rice irrigation in Asia), glaciation in northeastern Canada might have started long before the Industrial Revolution (Ruddiman, 2003a, 2007). However, the present interglacial may be much longer lasting than the past few warm periods because of (1) a rather different insolation pattern over the next 100 kyr due to the imprint of the 400-kyr eccentricity cycle in the Earth s orbit (see Fig. 3), and (2) the impact of relatively recent anthropogenic activities which have resulted in relatively large greenhouse gas concentrations in the atmosphere. With regard to (1), it has been argued that the best analogue for the present interglacial is MIS 11, which started around 410 kyr BP and lasted for about thirty thousand years (Fig. 1; EPICA Community Members, 2004). (However, Ruddiman (2007) has recently suggested La Société canadienne de météorologie et d océanographie

4 Glacial Inceptions: Past and Future / 319 enthic δ O( ) 18 MIS Age (kyr) Fig. 1 Time series of δ 18 O measurements taken from benthic foraminifera in an ocean sediment core from Ocean Drilling Program (ODP) Site 980 (55º29 N, 14º42 W, 2179 m) over the last 0.5 million years (adapted from Wang et al. (2002) by permission of the American Geophysical Union). Benthic data were obtained from Cibicidoides wuellerstorfi. This record is a proxy for global ice volume (increasing downward). The last interglacial period extended over the period kyr BP (before present; here present is defined as 1950), the Last Glacial Maximum (LGM) occurred at 21 kyr BP, and the present interglacial period (the Holocene) started at around 10 kyr BP. Peak glacial periods also occurred at approximately 135 kyr BP, 250 kyr BP, 340 kyr BP and 420 kyr BP, during MIS 6, 8, 10 and 12, respectively. MIS stands for the marine isotope stage. that the method used by the EPICA Community Members to estimate the length of the interglacial MIS 11 is flawed, and that a closer look at the δd data used by EPICA reveals that this interglacial probably lasted less than 10 kyr.) With regard to (2), several EMIC studies have shown that the long-term concentration of atmospheric CO 2 in the distant future is a crucial factor in determining when the next glacial inception might occur (e.g., Berger and Loutre, 2002; Archer and Ganopolski, 2005; Cochelin et al., 2006). The remainder of this paper is structured as follows. A brief overview of the Milankovitch theory of glacial inception is given in Section 2, and a review of some earlier work on simulations of the LGI is presented in Section 3. In Section 4, the McGill Paleoclimate Model (MPM) is described in the context of other EMICs, and in Section 5 simulations of the LGI with both the geophysical and green MPM are given. Possible scenarios for the next glacial inception are presented in Section 6, and some concluding remarks are given in Section 7. 2 On the theory of glacial inceptions The Milankovitch (1941) theory of long-term (multi-millennial) climate change states that a glacial inception occurs when the summer insolation at high northern latitudes decreases substantially and reaches a very low value. However, changes in the concentration of greenhouse gases in the atmosphere amplify the glacial inception process, According to Martin Claussen (personal communication, 2006), this theory is based on an earlier hypothesis published in Köppen and Wegener (1924). through different feedbacks (e.g., Gallée et al., 1992; Ruddiman, 2003b). The solar forcing at the top of the atmosphere slowly changes due to variations of three orbital parameters of the Earth s motion about the sun: (1) the eccentricity, with periodicities of around 100 and 400 kyr, (2) the obliquity, with a period of 41 kyr, and (3) the climatic precession (a measure of the Earth-sun distance during the summer solstice), with a dominant period of 23 kyr (see Milankovitch (1941) and the expanded explanation in Ruddiman (2001)). Milankovitch argued that the optimal conditions to enter a glaciation are a high climatic precession, a high eccentricity and a low obliquity (which collectively result in a low insolation in summer at high northern latitudes and a low seasonal contrast). Such conditions occurred around 116 kyr BP (see point A in Fig. 2), and as a consequence, the insolation in June at 62.5 N reached the very low value of about 440 W m 2 (see the lowest arrow in the left-hand side of Fig. 3) and a glacial inception occurred (Fig. 1). The evolution of the orbital parameters over the next 100 kyr does not replicate the above conditions for glacial inception because of the small eccentricity and weak variations in the climatic precession (which is modulated by the eccentricity) over this period (see left-hand side of Fig. 2). Since the eccentricity is currently near the end of a 400-kyr cycle, its value will be small for the next 100 kyr, as will the decreases in summer insolation at high northern latitudes (see arrows with question marks in Fig. 3). In fact, during the next 100 kyr, Fig. 3 shows that the next weak insolation of about 465 W m 2 will occur at 50 kyr AP (after present, defined here Canadian Meteorological and Oceanographic Society

5 320 / Lawrence A. Mysak Fig. 2 Past and future variations of the Earth s orbital parameters (as calculated by Berger (1978)) that affect long-term climate changes: The climatic precession (top), the obliquity (middle), and the eccentricity (bottom). Point A marks the beginning of the last glacial inception (LGI) during MIS 5d (see Fig. 1). Negative values on the time axis refer to times before 1950 and positive values refer to the future, after as after 1950). This insolation value is significantly larger than the low insolation values characteristic of the last four glacial inceptions (see arrows on the left-hand side of Fig. 3). Since the future summer insolation variations at high northern latitudes will have low amplitudes during the next 100 kyr, it is not obvious when the next glacial inception might occur in 50 kyr, 100 kyr or later? Because of our knowledge of past CO 2 concentrations in the atmosphere and their relation to the 100-kyr glacial-interglacial cycles (EPICA, 2004), it is highly likely that the future atmospheric CO 2 level will play an important role in determining the occurrence of the next glacial. Antarctic ice core records show that the atmospheric CO 2 level varied between approximately 180 and 280 ppm during the past half-million years (EPICA Community Members, 2004), with low values of around 180 ppm occurring during the peak glacials, and high values of around 280 ppm occurring during the interglacials. Since the Industrial Revolution, which started in the mid-eighteenth century, the CO 2 level has steadily increased to its current level of more than 380 ppm today (2007), and it will likely continue to increase to two-tothree times this value in the next 100 yr (IPCC, 2007). Recently, Archer and Ganopolski (2005) have argued that such large CO 2 concentrations will decrease very slowly over the next 100 kyr (because of the slow ocean uptake of atmospheric CO 2 ), and thus we shall experience an extremely long interglacial. Later in this paper, some of the simulations of Archer and Ganopolski (2005), along with those performed at McGill by Cochelin et al. (2006), will be reviewed with the aim of estimating how long the present interglacial will last. 3 Review of past simulations of the last glacial inception As mentioned previously, many of the earlier simulation attempts of the LGI (e.g., Royer et al., 1983; Rind et al., 1989; Oglesby, 1990) were made with seasonally varying atmospheric GCMs that were forced with insolation and radiative (atmospheric CO 2 ) fields appropriate for the timeslice at around 116 kyr BP. However, these simulations failed to produce substantial snow buildup at high northern latitudes and hence the studies pointed towards the importance of including amplifying feedbacks due to vegetation (e.g., Gallimore and Kutzbach, 1996; de Noblet et al., 1996), the oceans (Dong and Valdes, 1995; Khodri et al., 2001), sea ice (Yoshimori et La Société canadienne de météorologie et d océanographie

6 Glacial Inceptions: Past and Future / 321 W m 2 Fig. 3 Insolation at the top of the atmosphere (TOA) at 62.5 N in June, between 500 kyr BP (negative values) and 500 kyr AP (positive values), as calculated by Berger (1978). (Note: In this paper AP means after 1950.) A W marks a warm interglacial period, and an arrow (without a question mark) indicates the time of a past glacial inception. Starting from the left, the arrows correspond to MIS 11, 9, 7 and 5d shown in Fig. 1. The arrows with a question mark indicate the possible time of a future glacial inception. al., 2002), and the polar surface energy balance (Vettoretti and Peltier, 2003). Further, the sensitivity of LGI simulations to the initial size of the Greenland ice sheet has been investigated by Kubatzki et al. (2006). Khodri et al. (2001), in a coupled atmosphere-ocean GCM timeslice study of the LGI, showed that in their model ocean feedbacks lead to a cooling of the high northern latitudes, along with an increase in the moisture transport to the polar Canadian Meteorological and Oceanographic Society

7 322 / Lawrence A. Mysak latitudes. The latter leads to an increased delivery of snow to northern latitudes and the steady buildup of snow at around 70 N (Fig. 4). However, since the model of Khodri et al. (2001) is run for only 100 yr under the same seasonal cycle for the insolation at 115 kyr BP, the actual buildup of ice sheets over several millennia cannot be simulated. In reality, the LGI started at 119 kyr BP, and by 115 kyr BP substantial ice was present. Therefore, ideally one should start an LGI run at around 120 kyr BP and let it continue to 115 kyr BP and beyond. To carry out such simulations, long-term transient runs of climate system models are necessary, and the appropriate tool to carry out such runs is the EMIC, which generally includes most interactive components of the climate system. Among the first EMICs which coupled most components of the climate system is that of Gallée et al. (1991, 1992). In the 1991 paper, a 2-D (latitude-height) model of the northern hemisphere atmosphere (zonally averaged) was developed and coupled to an ocean mixed-layer, sea ice and land surface components. In Gallée et al. (1992), this four-component climate model was asynchronously coupled to a model of the three main northern ice sheets and the accompanying bedrock. This coupled climate-ice sheet model is often referred to as the 2-D Louvain-la-Neuve (LLN) EMIC. Starting at 120 kyr BP, with forcing consisting of the astronomically derived insolation (Berger, 1978) and atmospheric CO 2 concentration obtained from the Vostok ice core (Barnola et al., 1987), the model was able to simulate the rapid latitudinal growth of the North American and Eurasian ice sheets from 120 to 110 kyr BP, as well as the last glacial maximum ice-sheet volume at 19 kyr BP (see Fig. 7a in Gallée et al., 1992). The last two glacial cycles were also simulated with this model by Loutre and Berger (2000). Wang and Mysak (2002; hereafter referred to as WM2002) were the first to carry out transient run simulations of the LGI using a global five-component EMIC which included a zonally averaged latitude-depth model for the ocean s thermohaline circulation (THC) (Wright and Stocker, 1991). The other four components of this EMIC are the 2-D (latitude-longitude) dynamic ice-sheet model of Marshall and Clarke (1997), and the atmosphere, sea-ice, and land-surface models as described in Wang and Mysak (2000). Under Milankovitch forcing and Vostok-derived atmospheric CO 2 for the period 122 to 110 kyr BP, the model (called the geophysical MPM) was used to investigate the mechanisms involved in the LGI, including those associated with the THC in particular. A novel result found by WM2002 is that during the buildup of northern hemisphere ice sheets during the LGI, the THC also intensified, which allowed for large moisture transports from the oceans to the continents at high northern latititudes. This model will be described in more detail in Section 4, and the LGI simulations using the MPM will be presented in Section 5a. The earlier GCM timeslice studies that illustrated the importance of vegetation in ice-sheet growth during the LGI (e.g., de Noblet et al., 1996) motivated Crucifix and Loutre (2002), Kageyama et al. (2004), Calov et al. (2005a, 2005b), Wang et al. (2005) and Kubatzki et al. (2006) to conduct various EMIC transient LGI simulations that included a dynamic vegetation component. Crucifix and Loutre (2002) showed how vegetation works in synergy with snow cover and sea ice to produce inception conditions. Through the vegetation-albedo feedback mechanism (which is similar to the familiar icealbedo feedback, except that boreal vegetation is a highly energy-absorbing surface, whereas ice is a highly reflective surface), Kageyama et al. (2004) showed that vegetation changes over North America (the northern boreal forest gradually disappears before inception) amplify the insolationinduced cooling and initial ice-sheet buildup there. This process does not occur over Eurasia in their model because the initial climate there is warmer, and vegetation is farther away from the taiga-tundra threshold which must be crossed for glacial inception to occur. Further, if vegetation is fixed at interglacial conditions, inception does not occur anywhere (see Fig. 1b in Kageyama et al. (2004)). Interestingly enough, Kubatzki et al. (2006) found that the LGI does not occur for fixed Eemian vegetation, but does occur for fixed present-day vegetation. Consistent with Kageyama et al. (2004), Wang et al. (2005) found that due to the vegetation-albedo feedback, large ice-sheet buildup during the LGI occurred over North America. However, in contrast to Kageyama et al. (2004) but in agreement with Calov et al. (2005a), Wang et al. (2005) simulated small ice-sheet buildup over northwest Europe and eastern Siberia when vegetation is fully interactive in the model. Archer and Ganopolski (2005) have been able to successfully simulate the past five 100-kyr ice age cycles using the conceptual model of Paillard (1998) in which the threshold for glacial inception was made to depend on the CO 2 concentration as determined from a stability analysis (Calov and Ganopolski, 2005) performed with the CLIMBER-2 EMIC from Potsdam (Brovkin et al., 2002) coupled to the 3-D thermomechanical ice-sheet model Simulation Code for Polythermal Ice Sheets (SICOPOLIS) (Greve, 1997) (see lefthand side of Fig. 5). These cycles were forced by the astronomically derived insolation and Vostok-derived CO 2. However, to simulate possible future glaciations realistically, an atmosphere-ocean and seafloor carbon cycle model (Archer, 2005) was used to determine future atmospheric CO 2 levels resulting from 300, 1000 and 5000 GtC anthropogenic releases to the atmosphere, neglecting natural carbon cycle variability. The partitioning of a large release of anthropogenic carbon between the atmosphere and the CaCO 3 -buffered oceans is such that, in the absence of natural CO 2 forcing, approximately 7% of the anthropogenic CO 2 remains in the atmosphere 100 kyr after the perturbation (see upper right-hand panel of Fig. 5). Thus, it appears that the long-term future levels of atmospheric CO 2 will likely be well above the pre-industrial value of 280 ppm. The impact of this result on future glaciations will be discussed further in Section 6, where other scenarios for the next glacial, obtained by the McGill Earth System Modelling Group, will be presented. La Société canadienne de météorologie et d océanographie

8 Glacial Inceptions: Past and Future / 323 Fig. 4 Monthly time series (40-yr record) of snow depth at a glaciation-sensitive region in northern Canada (70 N, 80 W), as calculated by Khodri et al. (2001) from two 100-yr runs in a coupled Atmosphere-Ocean General Circulation Model (A-OGCM). Time starts from January of year 50 for the control (present-day) experiment (solid line) and for the 115 kyr BP (glaciation) experiment (dot-dashed line), in which there is a reduced insolation at high northern latitudes in summer (see Fig. 3). The snow depth is very stable in the control run, with no snow in summer, whereas the 115 kyr BP run shows an increase in snow depth associated with perennial snow cover beginning around year 66 (month 792). 4 The McGill Paleoclimate Model (MPM) The MPM is a 2.5-D EMIC that has interactive atmosphereland-sea ice-ocean-ice sheet components; its origin can be traced back to the zonally averaged coupled atmosphereocean model of Stocker et al. (1992) that was developed at McGill nearly 20 years ago. However, unlike the model of Stocker et al., the MPM includes a moisture balance model for the atmosphere, as well as land surface, ice-sheet and seaice components; further, all components are seasonally varying and the atmosphere is forced by astronomically derived insolation (Berger, 1978) and Vostok-derived atmospheric CO 2. In addition, the zonal wind stress is a specified forcing field over the oceans. We call the MPM a 2.5-D model because while it has a 2-D ocean component (for latitude and depth), the other components are sectorially averaged across each ocean and continent, the domains of which are shown in Fig. 6. Note that the model domain only extends from 75 S to 75 N; hence the Arctic Ocean, some northern parts of North America and Eurasia, and the Antarctic continent have been omitted. The consequences of these limitations on ice-sheet buildup will be discussed in the next section. The ice-sheet component in the MPM is the vertically integrated dynamic part of the 3-D model of Marshall and Clarke (1997). It has a latitude-longitude resolution of 0.5 by 0.5. The atmospheric component is the energy-moisture balance model of Fanning and Weaver (1996), with an improved water vapour-temperature feedback (see WM2002). Further, as described in WM2002, the atmospheric variables (surface air temperature (SAT), surface specific humidity and precipitation) are downscaled to 5 by 5 in the region 30 N to 75 N. For details on the sea-ice and land-surface components, see Wang and Mysak (2000). This geophysical MPM was extended by Wang et al. (2005a) to include a new land surface scheme (LSS) with vegetation dynamics; this new model has become known as the green MPM. The LSS is characterized by the following improvements over the original version in Wang and Mysak (2000): (1) parameterization of deciduous and evergreen trees by using the model s climatology and the output of the dynamic global Vegetation COntinuous DEscription (VECODE) model (Brovkin et al., 2002) which determines in each grid cell, the fractions covered by trees, grass and desert; (2) parameterization of tree leaf budburst and leaf drop; (3) parameterization of the seasonal cycle of the grass and tree leaf area indices; and (4) calculation of the land surface albedo by using vegetation-related parameters, snow depth and the model s climatology. In addition, a systematic parameterization of the solar energy disposition (Wang et al., 2004) is now included in the green MPM. The green MPM s simulation of the present-day climate compared with that in the geophysical MPM is much improved. In particular, the strong seasonality of terrestrial vegetation and the associated land surface albedo variations are in good agreement with several satellite observations of these quantities. It is shown that with the explicit representation of the vegetation-albedo feedback in the model, slow millennial-scale climate changes during the Holocene are particularly well simulated (Wang et al., 2005b). We shall see in Section 5b that this feedback also plays an important role in the LGI. A detailed discussion of EMICs is given in Claussen et al. (2002), where it is shown that this class of models lies in the middle of the broad spectrum of climate models, which extends from conceptual (box) models (e.g., Paillard, 1998) to comprehensive high-resolution coupled atmosphereocean GCMs (see Fig. 1 in Claussen et al.). Generally speaking, each EMIC falls into one of three categories: (1) a 2-D or 2.5-D model like the MPM which is based on a zonally averaged ocean THC; (2) a hybrid model in which the ocean Canadian Meteorological and Oceanographic Society

9 324 / Lawrence A. Mysak a b c d Fig. 5 Simulation by Archer and Ganopolski (2005), reproduced by permission of the American Geophysical Union, of the past five glacial-interglacial cycles (past 500 kyr) and of possible future interglacial-glacial cycles during the next 500 kyr. For the future runs, the CLIMBER-2/SICOPOLIS ice sheet model is coupled to an atmosphere-ocean and seafloor carbon cycle model (Archer, 2005). For the past, the model is driven by atmospheric CO 2 derived from the Vostok ice core (Petit et al., 1999) and insolation changes as calculated by Berger (1978) (see left-hand side of panels a and b). The green curves represent the natural evolution of climate (see panels c and d), and the blue, orange and red curves (on the right-hand side) represent, respectively, the results for short-term anthropogenic releases into the atmosphere of 300, 1000 and 5000 GtC. (a) Past and future p CO 2 of the atmosphere, according to, respectively, Petit et al. (1999) and the carbon cycle model of Archer (2005). (b) June insolation at 65 N normalized and expressed in units of the standard deviation, σ. 1 σ equals about 20 W m 2. The green, blue, orange and red lines are the values of the critical insolation, i 0, that triggers glacial inception. (c) The interglacial periods simulated by the model. (d) Global temperatures simulated by the model (green, blue, orange and red curves) and the past temperature as estimated for the Vostok ice core (black curve). component say, is a coarse-resolution 3-D GCM, that is coupled to a simpler (e.g., energy-moisture balance) atmospheric model; or (3) a coarse-resolution 3-D atmosphere-ocean model with sea-ice and land-surface representations in which many of the processes are simplified. In addition to the MPM, the Bern 2.5-D model, CLIMBER-2 (Potsdam) and MoBiDiC (Louvain-la-Neuve) fit into category (1), whereas the University of Victoria model fits into category (2). Among the models in category (3) are those from the Massachusetts Institute of Technology and the Russian Academy of Sciences, and also the EMIC EcBilt-CLIO (from Louvain-la-Neuve). For more details on these models, see Tables 1 and 2 in Claussen et al. (2002). A perusal of these tables indicates that these EMICs or their updates have interactive components for the complete Earth system, including the biosphere. Descriptions of the presentday versions of many EMICs, including some that were not included in Claussen et al. (2002), can be found on the following website: de/emics/toe_ pdf La Société canadienne de météorologie et d océanographie

10 Glacial Inceptions: Past and Future / 325 latitude (degrees) longitude (degrees) Fig. 6 Land-sea configuration for the MPM (the yellow grids correspond to Greenland). The north-south resolution in the model is 5º latitude, except across the equator, where it is 10º. From these model descriptions, it is clear that EMICs do have their limitations. For example, they cannot be used to study interannual climate variability associated with the El Niño Southern Oscillation (ENSO); this requires a high-resolution atmosphere-ocean model, especially in the tropics. However, they are an excellent tool to carry out simulations of many millennia of climate history and to investigate the interactions of as many components of the Earth system as possible in an efficient manner. Moreover, they can be used to explore, quite thoroughly, the parameter space of a model. Thus, they are more suitable for assessing uncertainty, which GCMs can do to a significantly lesser extent. Finally, from long transient runs of EMICs, we can identify interesting timeslices in the evolution of climate that can later be thoroughly investigated with GCMs. 5 Simulation of the LGI with the MPM a Simulations Without Vegetation During the initiation phase of the LGI, from 122 to 110 kyr BP, the northern North Atlantic, south of Iceland, was relatively warm at the surface (see Ruddiman and McIntyre (1979) and Fig. 7 (this paper), middle curve), likely because of an intensified THC (McManus et al., 2002). However, in the Norwegian Sea farther north, the sea surface temperature (SST) started to drop during MIS 5e at around 125 kyr BP and the surface waters cooled by 3 4ºC by 120 kyr (e.g., Cortijo et al., 1994). This cooling at around 120 kyr BP is consistent with earlier findings of Kellogg (1980). As the far northern parts of the main continents started to cool during the first half of this period due to reduced summer insolation (see Fig. 3 and also Fig. 11a in Section 5b), any moisture transported to these regions from the warm ocean around 50 60ºN would have fallen as snow and remained there year after year. Thus, as the climate continued to cool because of the orbital forcing, the accumulated snow would have led to rapid ice sheet growth through the ice-albedo feedback. Indeed, from Fig. 7 (lower curve) we can infer that by 110 kyr BP (during MIS 5d), a substantial amount of land ice formed. According to Lambeck and Chapell (2001), the global sea level had dropped about m by this time, which is equivalent to an ice volume in the range of km 3. Ice sheets with this volume would have become unstable near the margins of the North Atlantic, leading to large iceberg discharges and ice rafted debris (IRD) deposited in the deep ocean at around 107 kyr BP (see top curve in Fig. 7). Clearly, a measure of the success of any simulation of the LGI is whether large ice sheets in the volume range above can build up over the northern continents during a 5 10 kyr period after 120 kyr BP. Canadian Meteorological and Oceanographic Society

11 326 / Lawrence A. Mysak Neogloboqu pachyderma Cibicdoides wuellerstorfi Fig. 7 Paleoceanographic data taken from ODP site 980 in the northeast North Atlantic (55 29 N, W) (J.F. McManus, personal communication, 2002). Top curve: ice rafted debris (IRD); middle curve: proxy for SST as derived from δ 18 O measurements of planktonic foraminifera (Neogloboquadrina pachyderma s.); bottom curve: proxy for global ice volume (increasing downward) derived from δ 18 O measurements of benthic foraminifera (Cibicidoides wuellerstorfi). MIS stands for marine isotope stage. Table 1 shows the seven different runs that were made with the MPM in WM2002 to investigate the relative roles of various processes deemed to be important for early stages of the LGI and the subsequent ice sheet growth. In control run 1, the fully coupled MPM was integrated from 122 to 110 kyr BP with the radiative forcing shown in Fig. 1 of WM2002. (This forcing is also shown as the first 12 kyr of the time series in Fig. 11 in Section 5b.) Important model features included in this run are the elevation cooling effect of orography and a parameterization for the freezing of rain at high latitudes and refreezing of glacial meltwater (see WM2002 for details). In the other runs, 2 to 7, different constraints or effects were taken out of the model, until finally only Milankovitch forcing was used to drive the atmosphere-ice sheet model components. Figures 8a, 8b and 8c show, respectively, the time series of total, North American and Eurasian ice volumes for all the runs listed in Table 1. Clearly, the total ice-sheet growth (in red), starting around 119 kyr BP, is most rapid for run 1, with the total volume reaching about km 3 (Fig. 8a), which is about two-thirds of the observed value, as measured by sea level drop (Lambeck and Chapell, 2001). While this underestimation could be due to shortcomings of the model (e.g., the ice-sheet freezing and refreezing parameterization, the coarse resolution), it is most likely due to the limited domain of the model; north of 75 N there is no ice-sheet formation, in contrast to what is likely to have happened in reality. Fixing the freshwater flux into the ocean (run 2) or the SST (run 3) has a relatively small impact on the total ice volume growth, in agreement with Kageyama et al. (2004). However, neglecting the elevation effect of orography (run 4) or the freezing/refreezing parameterization (run 5) has a major impact on the growth. This first effect was also investigated by Kageyama et al. (2004), who found that it did not have a crucial effect on ice-sheet growth. Finally, in the Milankovitch run 7, the ice sheet growth is very small; this La Société canadienne de météorologie et d océanographie

12 Glacial Inceptions: Past and Future / 327 TABLE 1. Experimental design for simulations of the LGI with the MPM. For run 2, P E + R into the ocean is prescribed. When the SST is prescribed (run 3), the atmosphere-ocean heat and freshwater fluxes are fixed at their initial values. Refreezing means freezing of rain and refreezing of meltwater (reproduced from WM2002 by permission of the American Geophysical Union). Run Coupling CO 2 Mountain Freezing/Refreezing 1. Control run Fully coupled Vostok yes yes 2. Fixed freshwater flux P-E+R prescribed Vostok yes yes (into the ocean) 3. Fixed ocean SST prescribed Vostok yes yes 4. No mountain Fully coupled Vostok no yes 5. No freezing/refreezing Fully coupled Vostok yes no 6. No mountain and no Fully coupled Vostok no no freezing/refreezing 7. Milankovitch only SST prescribed 280 ppm no no Fig. 8 Simulated ice volume growths for the different runs described in Table 1: (a) total ice volume, (b) North American ice volume, and (c) Eurasian ice volume. (Reproduced from WM2002 by permission of the American Geophysical Union.) run helps to explain why running atmosphere-only GCMs for the LGI is generally not successful in simulating the LGI. Comparison of Figs 8b and 8c reveals that, in most runs, the ice volume growth is similar for both continents, which is at variance with the general belief that the ice sheets were larger over North America than over Eurasia during the last glacial (e.g., Turon, 1984). In Section 5b, we shall show that when an interactive vegetation component is included in the MPM, this discrepancy is removed. Figure 9a shows that during early glaciation, the THC in the control run is intensified until around 116 kyr BP; this is due to high latitude ocean cooling and reduced freshwater fluxes into the North Atlantic (Fig. 9b), which results in less buoyancy in the water column there and consequently Canadian Meteorological and Oceanographic Society

13 328 / Lawrence A. Mysak Flux Anomalies (Sv) Maximum THC Intensity (Sv) Fig. 9 The maximum THC intensities (20-yr means) (a), and freshwater flux anomalies (20-yr means) integrated over N (b) in the North Atlantic for the different runs described in Table 1. (Reproduced from WM2002 by permission of the American Geophysical Union.) enhanced North Atlantic Deepwater Formation (NADW). The strong THC (8 Sv above the interglacial maximum) increased the land-sea thermal contrast in the model and hence produced large moisture fluxes to the land, which is favourable for rapid ice-sheet growth. For run 2, the surface freshwater flux into the ocean was fixed, and the THC intensity increase during the first 5.7 kyr is only 4 Sv. This leads to a drop in SST there (compared to run 1) and extensive seaice formation in the NADW region just prior to 116 kyr BP. This results in a lower heat loss to the atmosphere and hence a drop in the THC intensity (Fig. 9a, green curve). Figure 10 shows snapshots of ice sheet distributions over the northern continents at three timeslices for the control run. At 120 kyr BP (Fig. 10a), ice first appears in the vicinity of the northern Laurentide, Scandinavian and Siberian ice-sheet regions. By 116 kyr BP (Fig. 10b), these ice sheets have expanded and new ice sheets have formed over Alaska and eastern Canada. By 110 kyr BP (Fig. 10c), thick ice sheets of order 3 km have formed over Alaska, eastern Canada and northeastern Europe. It is unlikely that such large sheets formed over Alaska during the LGI (A. Dyke, personal communication, 2002; W.F. Ruddiman, presonal communication, 2007), because of the large mountains in this region and the increased transport of latent and sensible heat there under 116 kyr BP orbital forcing that produced summer snow melt (Vettoretti and Peltier, 2003). Moreover, in contrast to observations (Turon, 1984), the model overestimates the volume of ice formed in Eurasia during the LGI. However, the simulation of ice in Siberia between 116 and 110 kyr BP is consistent with the oxygen isotope evidence of Siberian glaciation during MIS 5d presented by Karabanov et al. (1998). b Simulations With Vegetation In the green MPM, the dynamic vegetation model VECODE, developed for use in EMICs by Brovkin et al. (2002), has been interactively coupled to the geophysical MPM in order to incorporate the biogeophysical vegetation albedo feedback. In addition, as noted in Section 4, a new land surface scheme has been introduced (Wang et al., 2005a). Furthermore, in contrast to WM2002, the Greenland ice sheet is now explicitly resolved, being located in the western half of the North Atlantic as part of the ice sheet model, but attached to the North American continent (see yellow grid boxes in Fig. 6) for the purpose of coupling it to the other components of the MPM. The green MPM has been forced with variable insolation (Fig. 11a) and atmospheric CO 2 concentration (Fig. 11b) for the period from 122 to 80 kyr BP, which is 30 kyr longer than La Société canadienne de météorologie et d océanographie

14 Glacial Inceptions: Past and Future / 329 longtitude (degrees) Fig. 10 Ice sheet thickness distributions over North America and Eurasia for the control run 1 in Table 1 at 120 kyr BP (a), 116 kyr BP (b), and 110 kyr BP (c). (Reproduced from WM2002 by permission of the American Geophysical Union.) (a) Insolation in June at 62 5ºN Atmospheric CO (b) CO 2 concentration derived from Vostok data W m 2 Fig. 11 (a) Insolation at a high northern latitude in summer, as calculated by Berger (1978), and (b) Vostok-derived atmospheric CO 2 concentration taken from Banola et al. (1999), between 122 and 80 kyr BP. the run in WM2002. In Wang et al. (2005) results are presented for both a control run (in which the vegetation is interactive globally) and a number of sensitivity runs with vegetation fixed in different regions (see Table 1 in Wang et al. for details). Here, the focus will be mainly on describing the control run. Figure 12 shows the time series for the total ice volume (heavy green line), and also the ice volumes over North America (NA) and Eurasia (EUR). At 110 kyr BP, the Canadian Meteorological and Oceanographic Society

15 330 / Lawrence A. Mysak Sea level-equivalent ice volume from Lambeck and Chappell (2001) Total ice volume produced by the green MPM Ice Volume (x 10 6 km 3 ) Fig. 12 Ice volume growth simulated by the green MPM between 122 and 80 kyr BP. Heavy green curve: total ice volume over North America and Eurasia; light green curves: ice volume simulated over North America (NA) and Eurasia (EUR). Dotted red line: sea-level equivalent global ice volume estimated by Lambeck and Chappell (2001). (Reproduced from Cochelin (2004) by permission.) total ice volume is slightly smaller than that in the control run in the geophysical MPM (see Fig. 8a); however, now we observe that with interactive vegetation in the model, the ice sheet growth over North America is much larger than over Eurasia, a feature which is believed to have happened in reality (Turon, 1984). Further, we note that there is a small icesheet buildup over Eurasia at 110 kyr BP, which was not obtained by Kageyama et al. (2004). At the end of the run, on the other hand, the ice volumes over each continent are comparable, with the North American ice volume being slightly larger. Also, at 80 kyr BP, the total ice volume simulated is approximately equal to the observed volume, as estimated from sea level changes. The time series for the sea-level equivalent ice volume over the integration period (dashed red curve in Fig. 12) shows large fluctuations which may be due to massive iceberg discharges, a feature not simulated in the green MPM because of the lack of ice-sheet thermodynamics. The first large interruption in ice-sheet growth is presumably due to the substantial insolation increase after 115 kyr BP (see Fig. 11) and subsequent basal ice melt and rapid ice sliding, which led to the iceberg discharges. Figure 13 shows the simulated ice thickness distributions at six timeslices of the control run. By 116 kyr BP, permanent ice has appeared in Alaska, and the Laurentide, Scandinavian and Siberian regions; however, in contrast to the case in the geophysical MPM, there is much less ice in Eurasia (compare Figs 13b and 10b). Between 116 and 80 kyr BP, the ice sheets continue to expand longitudinally, towards the centre of the continents, as well as southward. At the end of the run (Fig. 13f), substantial ice sheets appear over Alaska, Canada and northwestern Europe. According to evidence from glacial deposits and striation patterns, the large simulated ice sheet over Alaska is unrealistic (W.F. Ruddiman, personal communication, 2007) Figure 14 illustrates the evolution of the northern tree and desert fractions in the control run (red curves). The simulated tree fraction in the control run follows closely the 23-kyr precessional cycle for the insolation (Fig. 11 a); the desert fraction changes are opposite to these insolation variations. Figure 14 also shows the results for two sensitivity experiments: the evolution of the above fractions for either fixed SAT (green curves) or precipitation (blue curves). Upon noting the similarity between the blue curves (with active SAT and fixed precipitation) and the red ones, we conclude that north of 60 N latitude the tree and desert fraction changes are driven predominantly by temperature changes. This is because the change in the number of growing degree days La Société canadienne de météorologie et d océanographie

16 Glacial Inceptions: Past and Future / 331 (a) 122 kyr BP (d) 100 kyr BP (b) 116 kyr BP (e) 90 kyr BP (c) 110 kyr BP (f) 80 kyr BP longitude (degrees) longitude (degrees) Fig. 13 Simulated ice-sheet thickness distributions in the control run of the green MPM at (a) 122 kyr BP, (b) 116 kyr BP, (c) 110 kyr PG, (d) 100 kyr BP, (e) 90 kyr BP, and (f) 80 kyr BP. (Reproduced from Wang et al. (2005) by permission of the American Geophysical Union.) Fig. 14 Tree fractions in the total land (including ice sheets) and (b) desert fractions in total land (including ice sheets), averaged between 60 and 75 N in the green MPM, for the control experiment (red), the experiment with fixed SAT (green) and the experiment with fixed precipitation (blue) in the vegetation component. (Reproduced from Wang et al. (2005) by permission of the American Geophysical Union.) drives the high latitude vegetation changes (Wang et al., 2005a). An increase in the insolation during the warm season in this region increases both the length and the temperature of this season, which favours tree growth. Figure 15 shows, for the control run, the modelled tree fraction distribution at three timeslices (122, 100 and 80 kyr BP). By comparing these plots with the simultaneous ice sheet distributions (in Figs 13a, 13d and 13f), we note that the Canadian Meteorological and Oceanographic Society

17 332 / Lawrence A. Mysak latitude (degrees) latitude (degrees) latitude (degrees) longitude (degrees) Fig. 15 Tree fractions for the entire model area in the control run of the green MPM at (a) 122 kyr BP, (b) 100 kyr BP, and (c) 80 kyr BP. Greenland is attached to the North American continent for plotting purposes. (Reproduced from Wang et al. (2005) by permission of the American Geophysical Union.) trees progressively disappear and the ice sheets build up and give way to desert in the high northern latitudes. The deserts (not shown) therefore expand to the places where the ice sheets are located. At 80 kyr BP, there are almost no trees anywhere in North America between 60 and 75 N due to the expansion of the ice sheets. The same is true in Eurasia between 65 and 75 N. In high northern latitudes, the treeline has shifted southward by 5 to 10 over the course of the simulation. In view of this, the tree fraction averaged between 60 and 75 N (see red curve in Fig. 14a) has substantially decreased between 122 and 80 kyr BP, whereas the desert fraction has greatly increased in this region (see red curve in Fig. 14b). These changes in vegetation and the associated changes in surface albedo contribute to the expansion of the ice sheets, owing to the positive vegetation-albedo and icealbedo feedbacks. 6 Simulation of the next (possible) glaciation Loutre and Berger (2000) ran the 2-D LLN hemispheric model for the next 130 kyr under orbital forcing and various constant concentrations of CO 2 ranging from 210 to 290 ppm. The red dotted ice volume curve in the bottom panel of Fig. 16, taken from the summary paper by Berger and Loutre (2002), shows that a glacial inception would occur immediately for the lowest value of CO 2 in the above range. However, for a future natural CO 2 variation similar to the variations of the past 130 kyr (as seen in the Vostok ice core), the present interglacial would last for at least another 50 kyr (see solid curve in bottom panel of Fig. 16). This is because of the small variations in the high northern latitude summer insolation for the next 50 kyr (see middle panel of Fig. 16). Upon introducing a global warming scenario starting at time La Société canadienne de météorologie et d océanographie

18 Glacial Inceptions: Past and Future / 333 Insolation (W m 2 ) Fig. 16 Long-term variations of eccentricity (top), June insolation at 65 N (middle), and simulated northern hemisphere ice volume (increasing downward) (bottom) from 200 kyr BP to 130 kyr AP (the present is defined as 1950). For the future simulations, three CO 2 scenarios were used: last glacial-interglacial values, a human-induced (global warming) concentration which peaks at 750 ppm at 200 yr AP, and a constant concentration of 210 ppm. These scenarios produced the ice volumes indicated by the solid, red dashed and red dotted curves, respectively. Simulation results from Loutre and Berger (2000); eccentricity and insolation from Berger (1978). zero (1950) in Fig. 16, in which the CO 2 rises linearly from 280 to 750 ppm over the next 200 yr and then slowly drops back down to the pre-industrial value of 280 ppm over the next 800 yr, the Greenland ice sheet melts initially but then reforms. The model in this scenario forecasts the end of the present interglacial also in about 50 kyr (see dashed red curve in the bottom panel of Fig. 16), which is at the same time as the present interglacial would end in the natural CO 2 run (solid curve). In Cochelin et al. (2006), experiments similar to those in Loutre and Berger (2000) were performed with the (global) green MPM, which was integrated forward in time (starting at 1950) for the next 100 kyr under orbital forcing (Berger, 1978) and a variety of CO 2 scenarios. The first set of simulations was run under various constant atmospheric CO 2 levels. In the second set of simulations, the CO 2 level rapidly increases over the first 350 yr to 1200 ppm in year 2300 and then slowly decreases over the next 850 yr until it stabilizes at various levels at 1.2 kyr AP (see inset in Fig. 20). For the remaining 98.8 kyr, the atmospheric CO 2 level remains constant. This variation in CO 2 represents the inclusion of a large global warming episode superimposed on the constant CO 2 scenario. As will be discussed later in this section, the assumption that atmospheric CO 2 will ultimately return (after 1 to 2 kyr) to around a pre-industrial level after a large input of carbon into the atmosphere may not be realistic (Archer, 2005). Accordingly, if larger levels of CO 2 remain in the atmosphere for thousands of years, the current interglacial could last a very long time (Archer and Ganopolski, 2005). Figure 17 shows the time series of ice volume over North America for the next 100 kyr in the first set of experiments. Figure 18 illustrates the evolution of the maximum intensity of the THC for these experiments, and Fig. 19 portrays the tree and desert fraction changes averaged over the high northern latitudes. From Fig. 17, we observe that, depending on the CO 2 level, there are three possible types of evolution for the ice volume: an imminent glacial inception, a glacial inception in 50 kyr, or no glacial inception during the next 100 kyr. Mathematically speaking, the climate system model passes through two thresholds for glaciation/no glacial inception as the CO 2 concentration is increased. For CO 2 levels less than or equal to 270 ppm, the climate enters into a glacial period quite quickly. This general result is consistent with that of Berger and Loutre (2002; see Fig. 16). Ice starts to build up in the west of the northern latitude region of North America and then slowly expands eastward and southward (figure not shown). We note that the Canadian Meteorological and Oceanographic Society